PREDICTION OF AFLATOXIN RISK IN CORN FEED BASED ON PHYSICAL QUALITY PARAMETERS: DEVELOPMENT AND VALIDATION OF ORDINAL MODELS

Abstract

Aflatoxin contamination in feed corn is a critical concern, especially in tropical regions where postharvest moisture and storage conditions support fungal growth.This study developed and validated an ordinal logistic regression model to predict the risk of aflatoxin contamination using corn physical quality parameters.A total of 384 valid observations were used for model development.Aflatoxin was classified into three ordinal risk categories, while moisture content, damaged kernels, broken kernels, moldy kernels, and foreign materials were classified into ordinal quality classes based on adjusted feed -corn quality criteria.The initial model included all physical quality parameters, and predictor selection was performed using Wald statistics.The final model retained moisture content, broken kernels, and moldy kernels as significant predictors, whereas damaged kernels and foreign materials were excluded.Model fitting indicated that the final model significantly improved prediction compared with the intercept-only model (P<0.001).Goodness-of-fit statistics were non-significant, and the Nagelkerke pseudo-R² reached 0.519, indicating adequate explanatory capability.Model validation with 105 independent samples yielded an overall classification accuracy of 73.33%, with high class-specific accuracy in the low-risk and high-risk categories.The moderate Cohen's Kappa value (κ = 0.597) indicated meaningful agreement between actual and predicted risk classes beyond chance.Progressive validation across increasing sample sizes (N = 10-200) showed stable performance, with accuracy ranging from 74.50 to 83.20% and Kappa values approaching a stable moderate-agreement range.These findings suggest that ordinal logistic regression can serve as an early-warning tool for screening aflatoxin risk in feed corn based on physical quality indicators.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint